Causal models to leverage time series data: an application on COVID-19 monitoring by wastewater sampling
For COVID-19 monitoring, detecting viral RNA load in wastewater samples has been suggested as a valid and cost-effective alternative to personal testing. However, the current statistical methods struggle to quantify the associated shedding population. They also neglect other information of epidemiological interest like reproduction number or counterfactual analysis, limiting widespread applications.
In this talk, instead, I will introduce a causal-based model that combines epidemiological dynamics and wastewater data. It is based on the Extended Kalman filter state-estimate procedure and is informed by a complex SEIR model, which effectively reproduces the wastewater data collected in Luxembourg.
The model quantitatively estimates COVID-19 prevalence in a community and forecasts epidemic evolution. In addition, it allows exploring plausible scenarios to inform decision-makers. Firstly, this approach is practically relevant for continuous and non-invasive surveillance of COVID-19. Moreover, it illustrates the benefits of complementing statistical analysis with causal models, which allow better explainability and enables further analysis.